hey, i'm
Samuel
Jayasingh
ai/ml engineer building intelligent systems.
currently @ spritle software · chennai, india.
$ cat ./about.md
i'm a junior software engineer specialising in artificial intelligence at spritle software, where i research and develop generative AI models — transformers, diffusion models, gans, and vaes.
my work spans llm-powered applications, multi-agent orchestration systems, rag pipelines, training language models from scratch, and production-grade ml — from industrial failure prediction to ai chatbots and image generation platforms.
b.tech in artificial intelligence and data science · rajalakshmi institute of technology · cgpa 8.0/10.
technologies
$ ls -la ./experiences
Jan 2026 – Present
Junior Software Engineer, AI
Spritle Software · Chennai
- researching and developing generative ai models: transformers, diffusion, gans, vaes
- building llm-powered applications and ml pipelines with pytorch, tensorflow, hugging face
- optimising model performance and developing production-grade ai systems
Oct 2025 – Jan 2026
Software Engineer Intern, AI
Spritle Software · Chennai
- designed agentic ai architectures evolving from single-agent to multi-agent orchestration
- implemented rag pipelines using fastapi, chromadb, and async embedding workflows
- built api toolkits and postgresql streaming query tools using claude sdk and mcp servers
- developed automated vulnerability scanning and remediation workflows
Jul – Dec 2024
AI Automation Intern
Vleafy Technologies · Chennai
- built conversational ai chatbots using llms and nlp for customer interaction
- automated backend workflows with python and gupshup apis
- integrated ai modules into full-stack web apps via restful services
Jul 2023 – Jan 2024
Data Analyst & ML Intern
ZF Group – WABCO India · Chennai
- designed a dynamic bus tracking system using mapping apis
- processed and engineered features from large-scale route data
- developed automated python dashboards for monitoring system metrics
$ ./terminal.sh --interactive
$ cat projects.json
a 351m-parameter gpt trained from scratch on 5b tokens, then reasoning-tuned with sft + grpo (rlvr) on math — full pretrain → sft → rlvr pipeline, end to end on a single amd mi300x, for under $100 total.
python · pytorch · rocm · grpo · hugging face
local-first agentos for minimal hardware — routes everyday queries to a free local gemma 4 model and escalates coding/reasoning work to cloud tiers, with multi-agent dispatch, persistent memory, cron jobs, and crash-recoverable event delivery.
python · fastapi · litellm · docker · gemma · fireworks ai
gpt-2 small (124m) trained from scratch on openwebtext — a file-per-concern reimplementation with flash attention, torch.compile, and bf16. 5k iterations on a single l4 gpu, val perplexity 27.3 against ~22.4 for the fully-trained original.
python · pytorch · flash attention · torch.compile · hugging face
fine-tuned qwen3.5-0.8b with lora on the llm-lat/harmful-dataset to strip its refusal behaviour — a red-teaming artifact for studying how fragile alignment is under targeted fine-tuning. held-out test loss 1.345, perplexity 3.84, on ~2 minutes of t4 compute.
python · lora / qlora · unsloth · hugging face · pytorch
mcp server exposing google stitch's ui generation api as structured tools — create projects, generate screens from natural language prompts, manage design systems, and produce design variants. lets claude and other ai clients drive interface creation conversationally.
python · mcp · google stitch api · fastmcp
mcp server exposing 40+ gitlab operations as structured tools for ai clients like claude and cursor — repo management, branch protection, merge requests, issue tracking, ci/cd log retrieval, and cross-project search. includes a standalone gemini-powered autonomous agent.
python · mcp · gitlab api · pydantic · docker
end-to-end ml pipeline for predicting industrial equipment failures from sensor telemetry. uses autoencoder-based anomaly detection, feature engineering on time-series data, apache airflow for orchestration, and mlflow for experiment tracking and model versioning.
python · autoencoders · apache airflow · mlflow · pytorch
agentic cli tool for automated application and infrastructure security analysis. autonomous agents handle threat modelling, cve mapping, and remediation suggestion across codebases and deployment configs.
python · llms · security · cli
$ ls -la ./blog
a reflection on the law of sowing and reaping, and why repentance — not effort — is what actually changes the harvest.
pretraining a 351m gpt from scratch, then reasoning-tuning it with sft and grpo (rlvr) on math — the pipeline that turns a fluent autocomplete model into one that gets rewarded for being correct, end to end on a single gpu for under $100.
training gpt-2 small from scratch on openwebtext, file by file — the architecture, the data pipeline trick that avoided a 54gb disk problem, and what the loss curves actually said.
a full walkthrough of the transformer machinery inside modern llms — tokens, embeddings, positional encoding, attention, feed-forward networks, the residual stream, and next-token prediction — with diagrams for each piece.
lora-finetuning qwen3.5-0.8b to strip its refusals on a free colab gpu, what it revealed about how fragile alignment really is, and why i published the result anyway.
a personal reflection on agape love, christian purpose, and what it means to live with faith at the centre — not just on sundays.
a practical look at building production systems with the anthropic claude sdk — tool use, streaming, mcp servers, and the projects i've shipped with it.
how i built an mcp server to expose 40+ gitlab operations as structured tools for ai clients — the protocol, the architecture, and what surprised me.
$ ./contact.sh
let's build
something.
always open to interesting problems, collaborations, and new opportunities. whether it's ai, full-stack, or something in between — reach out. also available for technical consultation and select project-based work.
contact@samueljayasingh.in ↗ book a consultation call ↗